April, 2012

TheMReport — News and strategies for the evolving mortgage marketplace.

Issue link:

Contents of this Issue


Page 64 of 82

FEATURE ANALYTICS of the third-party verification process, as well as compliance guidelines and regulations. "Lenders are seeking a risk analysis that constantly updates and changes as the loan makes its way through the loan process," said Russ Donnan, CIO at Loveland, Colorado-based Kroll Factual Data. "The result is an overall risk score, not a static risk score, set at application and then again at the pre-funding stage." In the past, predictive modeling typically targeted pricing and product eligibility efficiencies, and the mortgage industry responded with vendors providing services that addressed those needs. In today's very different market, lenders are trying to find ways to streamline their operations, ensure compliance and data integrity, and, of course, address the needs of all their stakeholders. "Today, however, the efficiency lenders are seeking not only encompasses speed, but also a heightened awareness of qual- ity as they benchmark them- selves," said Kelli Himebaugh of Southfield, Michigan-based Mortgage Builder Software. Roadblocks and Remedies In a rapidly changing industry with so many complex variables and different geographical influ- ences, it's no surprise that pre- dictive models have inefficiencies or outright flaws. No simple if-then statements here; perma- nence gives way to permutations in the statistical realm of the single-family mortgage industry. Interthinx's Smith cited roll- rate or transition models as a prime example. Though popular for capturing information about the delinquency status of a loan and using the data to estimate the likelihood of a change in that status, these models typically are unable to incorporate updated information about the loan that influences borrower behavior. "Just as a map does not equal the land it describes, models do not fully characterize the systems they describe," Smith said. "Users need to understand the strengths and weaknesses of their models." The biggest issue in creating Predictive Modeling From Loan Origination to Securitization R MBS bond valuation capabilities have advanced significantly since the economic downturn due to an increased focus on more granular data and analytics systems. The much- needed evolution has helped upgrade the securitization infrastructure and thereby raise the level of trust by all relevant participants in the process. According to Denver-based BlackBox Logic, some of the de- ficiencies in pre-recession RMBS analysis were: 1. Market assumptions about the Home Price Index tended to be too optimistic, and as a result, defaults and loss totals were underestimated. 2. Models often based their analysis on summary data rather 64 | THE M REPORT than performing analysis at a granular enough level. This was because the workflow and data necessary to target specific collateral sets backing the hundred thousands of bonds issued in the RMBS market were cumbersome and mechanically insurmountable. 3. Loan-level data in the non-conforming space was expensive, not of an extremely high quality, and therefore generally unavailable to market participants. The vast amounts of data could not be easily stored or manipulated, so the quality and timing of the downstream analysis suffered. Companies like BlackBox believe that past loan performance can be a good indicator of future performance as long as the analysis utilizes a comprehensive, granular, loan-level data set; cash flow predictions are targeted at specific loan collateral; and participants utilize the newly advanced analytic tools available in the industry today. Worthwhile data helps users fully understand how historic cash flows at the loan and bond levels were derived, which aids investors, traders, and researchers in calibrating their models to gain an in-depth understanding of their target assets. "Our clients can then take that cleansed information and the graphic displays describing the unscheduled cash flows of the loan collateral—i.e., voluntary prepayments, delinquencies, loss timing on foreclosed properties, loss severities generated from sales out of bank REO portfolios, effective mortgage models is obtaining accurate data sets that allow for predictive capabilities, Balkan said. Missing or inaccurate consumer data are frequent stumbling blocks for the research departments of mortgage companies and other financial institutions. "Also since predictive models are built on historical data sets, the constant change in the economic environment impacts the predictive power of these models," Balkan added. BlackBox Logic does not perform predictive analytics, but the Denver-based company does supply high-quality, granular residential mortgage loan-level data sets. Also, its Crystal Logic tool enables modelers to run historical time-series analysis, loan modifications, etc.—and integrate those data points into their predictive models," said Larry Barnett, BlackBox principal. "Due to a much better understanding of the setup data and more comprehensive information on subsequent loan-level performance, the predictive models that correlate loan attributes and historic behaviors to enable future cash flow predictions are now more accurate." Thanks to the comprehensive loan-level coverage, data integration, and automation, loan originators to bond investors can benefit from detailed cash flow predictions while saving time and energy. "Unfortunately, the historic ties between the front-end originators and the capital markets groups have been somewhat disaggregated due to the lack of securitization outside of the GSEs," Barnett said. "This has tended to slow this process. These improvements will eventually be incorporated into the residential mortgage and RMBS arenas." SECONDARY MARKET ANALYTICS SERVICING ORIGINATION

Articles in this issue

Archives of this issue

view archives of TheMReport - April, 2012